Matthew R. Huska
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Matthew R. Huska has a total of 225 co-authors. Here are the top co-authors by number of shared publications:
- Miguel A. Andrade-Navarro
Johannes Gutenberg University of Mainz
8 shared publications - Annalisa Marsico
Max Planck Institute for Molecular Genetics
3 shared publications - Gareth A. Palidwor
Ottawa Hospital Research Institute, Ottawa, ON K1R 6M1, Canada
3 shared publications Martin Vingron
Max Planck Institute for Molecular Genetics, Ihnestrasse 63-73 14195, Berlin, Germany
3 shared publicationsAnna Ramisch
Computational Molecular Biology, Max Planck Institute for Molecular Genetics, Berlin, Berlin, Germany
2 shared publications- Christopher J. Porter
Ottawa Bioinformatics Core Facility, Ottawa Hospital Research Institute, Ottawa, Canada
2 shared publications
Latest Publications
- Top Cited PapersOpen Access
- 9 February 2026
- Published by Springer Nature in Nature Methods
- Vol. 23 (3) , 653-662
Addressing pandemic-wide systematic errors in the SARS-CoV-2 phylogeny
The majority of SARS-CoV-2 genomes obtained during the pandemic were derived by amplifying overlapping windows of the genome (‘tiled amplicons’), reconstructing their sequences and fitting them together. This leads to systematic errors in genomes unless the software is both aware of the amplicon scheme and of the error modes of amplicon sequencing. Additionally, over time, amplicon schemes need to be updated as new mutations in the virus interfere with the primer binding sites at the end of amplicons. Thus, waves of variants swept the world during the pandemic and were followed by waves of systematic errors in the genomes, which had significant impacts on the inferred phylogenetic tree. Here we reconstruct the genomes from all public data as of June 2024 using an assembly tool called Viridian ( https://github.com/iqbal-lab-org/viridian ), developed to rigorously process amplicon sequence data. With these high-quality consensus sequences we provide a global phylogenetic tree of 4,471,579 samples, viewable at https://viridian.taxonium.org . We provide simulation and empirical validation of the methodology, and quantify the improvement in the phylogeny.
- 9 February 2026
- Published by Springer Nature in Nature Methods
- Vol. 23 (3) , 653-662
Top Cited PapersOpen AccessAddressing pandemic-wide systematic errors in the SARS-CoV-2 phylogeny
The majority of SARS-CoV-2 genomes obtained during the pandemic were derived by amplifying overlapping windows of the genome (‘tiled amplicons’), reconstructing their sequences and fitting them together. This leads to systematic errors in genomes unless the software is both aware of the amplicon scheme and of the error modes of amplicon sequencing. Additionally, over time, amplicon schemes need to be updated as new mutations in the virus interfere with the primer binding sites at the end of amplicons. Thus, waves of variants swept the world during the pandemic and were followed by waves of systematic errors in the genomes, which had significant impacts on the inferred phylogenetic tree. Here we reconstruct the genomes from all public data as of June 2024 using an assembly tool called Viridian ( https://github.com/iqbal-lab-org/viridian ), developed to rigorously process amplicon sequence data. With these high-quality consensus sequences we provide a global phylogenetic tree of 4,471,579 samples, viewable at https://viridian.taxonium.org . We provide simulation and empirical validation of the methodology, and quantify the improvement in the phylogeny.
- Open Access
Targeted decontamination of sequencing data with CLEAN
Background Many biological and medical questions are answered based on the analysis of sequence data. However, we can find contaminations, artificial spike-ins, and overrepresented rRNA sequences in various read collections and assemblies; complicating data analysis and making interpretation difficult. In particular, spike-ins used as controls, such as those known from Illumina (PhiX phage) or Nanopore data (DNA CS lambda phage, yeast enolase ENO2), are often not considered as contaminants and also not appropriately removed during bioinformatics analyses. Findings To address this, we developed CLEAN, a pipeline to remove unwanted sequence data from both long and short read sequencing techniques from a wide range of use cases. While focusing on Illumina and Nanopore data and removing of their technology-specific control sequences, the pipeline can also be used for everyday tasks, such as host decontamination of metagenomic reads and assemblies, or the removal of rRNA from RNA-Seq data. The results are the purified sequences and the sequences identified as contaminated with statistics summarized in an HTML report. Conclusions The decontaminated output files can be used directly in subsequent analyses, resulting in faster computations and improved results. Although decontamination is a task that seems mundane, many contaminants are routinely overlooked, cleaned by steps that are not fully reproducible or difficult to trace by the user. CLEAN will facilitate reproducible, platform-independent data analysis in genomics and transcriptomics and is freely available at https://github.com/hoelzer/clean under a BSD3 license.
Open AccessTargeted decontamination of sequencing data with CLEAN
Background Many biological and medical questions are answered based on the analysis of sequence data. However, we can find contaminations, artificial spike-ins, and overrepresented rRNA sequences in various read collections and assemblies; complicating data analysis and making interpretation difficult. In particular, spike-ins used as controls, such as those known from Illumina (PhiX phage) or Nanopore data (DNA CS lambda phage, yeast enolase ENO2), are often not considered as contaminants and also not appropriately removed during bioinformatics analyses. Findings To address this, we developed CLEAN, a pipeline to remove unwanted sequence data from both long and short read sequencing techniques from a wide range of use cases. While focusing on Illumina and Nanopore data and removing of their technology-specific control sequences, the pipeline can also be used for everyday tasks, such as host decontamination of metagenomic reads and assemblies, or the removal of rRNA from RNA-Seq data. The results are the purified sequences and the sequences identified as contaminated with statistics summarized in an HTML report. Conclusions The decontaminated output files can be used directly in subsequent analyses, resulting in faster computations and improved results. Although decontamination is a task that seems mundane, many contaminants are routinely overlooked, cleaned by steps that are not fully reproducible or difficult to trace by the user. CLEAN will facilitate reproducible, platform-independent data analysis in genomics and transcriptomics and is freely available at https://github.com/hoelzer/clean under a BSD3 license.
- Open Access
Refphase: Multi-sample reference phasing reveals haplotype-specific copy number heterogeneity
Most computational methods that infer somatic copy number alterations (SCNAs) from bulk sequencing of DNA analyse tumour samples individually. However, the sequencing of multiple tumour samples from a patient’s disease is an increasingly common practice. We introduce Refphase, an algorithm that leverages this multi-sampling approach to infer haplotype-specific copy numbers through multi-sample reference phasing. We demonstrate Refphase’s ability to infer haplotype-specific SCNAs and characterise their intra-tumour heterogeneity, to uncover previously undetected allelic imbalance in low purity samples, and to identify parallel evolution in the context of whole genome doubling in a pan-cancer cohort of 336 samples from 99 tumours.
Open AccessRefphase: Multi-sample reference phasing reveals haplotype-specific copy number heterogeneity
Most computational methods that infer somatic copy number alterations (SCNAs) from bulk sequencing of DNA analyse tumour samples individually. However, the sequencing of multiple tumour samples from a patient’s disease is an increasingly common practice. We introduce Refphase, an algorithm that leverages this multi-sampling approach to infer haplotype-specific copy numbers through multi-sample reference phasing. We demonstrate Refphase’s ability to infer haplotype-specific SCNAs and characterise their intra-tumour heterogeneity, to uncover previously undetected allelic imbalance in low purity samples, and to identify parallel evolution in the context of whole genome doubling in a pan-cancer cohort of 336 samples from 99 tumours.
- Open Access
Rise and Fall of SARS-CoV-2 Lineage A.27 in Germany
Here, we report on the increasing frequency of the SARS-CoV-2 lineage A.27 in Germany during the first months of 2021. Genomic surveillance identified 710 A.27 genomes in Germany as of 2 May 2021, with a vast majority identified in laboratories from a single German state (Baden-Wuerttemberg, n = 572; 80.5%). Baden-Wuerttemberg is located near the border with France, from where most A.27 sequences were entered into public databases until May 2021. The first appearance of this lineage based on sequencing in a laboratory in Baden-Wuerttemberg can be dated to early January ’21. From then on, the relative abundance of A.27 increased until the end of February but has since declined—meanwhile, the abundance of B.1.1.7 increased in the region. The A.27 lineage shows a mutational pattern typical of VOIs/VOCs, including an accumulation of amino acid substitutions in the Spike glycoprotein. Among those, L18F, L452R and N501Y are located in the epitope regions of the N-terminal- (NTD) or receptor binding domain (RBD) and have been suggested to result in immune escape and higher transmissibility. In addition, A.27 does not show the D614G mutation typical for all VOIs/VOCs from the B lineage. Overall, A.27 should continue to be monitored nationally and internationally, even though the observed trend in Germany was initially displaced by B.1.1.7 (Alpha), while now B.1.617.2 (Delta) is on the rise.
Open AccessRise and Fall of SARS-CoV-2 Lineage A.27 in Germany
Here, we report on the increasing frequency of the SARS-CoV-2 lineage A.27 in Germany during the first months of 2021. Genomic surveillance identified 710 A.27 genomes in Germany as of 2 May 2021, with a vast majority identified in laboratories from a single German state (Baden-Wuerttemberg, n = 572; 80.5%). Baden-Wuerttemberg is located near the border with France, from where most A.27 sequences were entered into public databases until May 2021. The first appearance of this lineage based on sequencing in a laboratory in Baden-Wuerttemberg can be dated to early January ’21. From then on, the relative abundance of A.27 increased until the end of February but has since declined—meanwhile, the abundance of B.1.1.7 increased in the region. The A.27 lineage shows a mutational pattern typical of VOIs/VOCs, including an accumulation of amino acid substitutions in the Spike glycoprotein. Among those, L18F, L452R and N501Y are located in the epitope regions of the N-terminal- (NTD) or receptor binding domain (RBD) and have been suggested to result in immune escape and higher transmissibility. In addition, A.27 does not show the D614G mutation typical for all VOIs/VOCs from the B lineage. Overall, A.27 should continue to be monitored nationally and internationally, even though the observed trend in Germany was initially displaced by B.1.1.7 (Alpha), while now B.1.617.2 (Delta) is on the rise.
- 1 January 2019
- Published by Elsevier in Cell Chemical Biology
- Vol. 26 (1) , 109-120.e7
The Anti-amyloid Compound DO1 Decreases Plaque Pathology and Neuroinflammation-Related Expression Changes in 5xFAD Transgenic Mice
- 1 January 2019
- Published by Elsevier in Cell Chemical Biology
- Vol. 26 (1) , 109-120.e7
The Anti-amyloid Compound DO1 Decreases Plaque Pathology and Neuroinflammation-Related Expression Changes in 5xFAD Transgenic Mice
- Open Access
- 16 December 2016
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 12 (12) , e1005249
Improved Prediction of Non-methylated Islands in Vertebrates Highlights Different Characteristic Sequence Patterns
Non-methylated islands (NMIs) of DNA are genomic regions that are important for gene regulation and development. A recent study of genome-wide non-methylation data in vertebrates by Long et al. (eLife 2013;2:e00348) has shown that many experimentally identified non-methylated regions do not overlap with classically defined CpG islands which are computationally predicted using simple DNA sequence features. This is especially true in cold-blooded vertebrates such as Danio rerio (zebrafish). In order to investigate how predictive DNA sequence is of a region’s methylation status, we applied a supervised learning approach using a spectrum kernel support vector machine, to see if a more complex model and supervised learning can be used to improve non-methylated island prediction and to understand the sequence properties of these regions. We demonstrate that DNA sequence is highly predictive of methylation status, and that in contrast to existing CpG island prediction methods our method is able to provide more useful predictions of NMIs genome-wide in all vertebrate organisms that were studied. Our results also show that in cold-blooded vertebrates (Anolis carolinensis, Xenopus tropicalis and Danio rerio) where genome-wide classical CpG island predictions consist primarily of false positives, longer primarily AT-rich DNA sequence features are able to identify these regions much more accurately. DNA methylation is one of the most heavily studied epigenetic modifications, with active research into the topic taking place for over 30 years. Over that time a relationship between non-methylated stretches of genomic DNA and the regulation of nearby genes has been identified, and it was noted that approximately half of all human genes have an unexpectedly high number of unmethylated CpG dinucleotides in their promoters. Nevertheless, until relatively recently we haven’t been able to directly measure methylation gnome-wide, leading to the widespread use of CpG-rich regions (or CpG Islands) as a proxy for unmethylated genomic regions. Fortunately, recently developed experiments finally allow us to look at DNA methylation genome-wide, and have shown that CpG islands do not overlap true non-methylated islands to the extent we would have hoped, especially in cold-blooded vertebrates. We show that by using these experimentally determined non-methylated regions as training data, we are able to improve the prediction of non-methylated regions genome-wide in comparison to classical CpG island prediction methods, in both warm-blooded and cold-blooded vertebrates. Additionally, we show that while CpG content is an important predictor of methylation status in warm-blooded vertebrates, in cold-blooded vertebrates higher predictive accuracy can be achieved using longer stretches of relatively AT-rich DNA sequence.
- 16 December 2016
- Published by Public Library of Science (PLoS) in PLOS Computational Biology
- Vol. 12 (12) , e1005249
Open AccessImproved Prediction of Non-methylated Islands in Vertebrates Highlights Different Characteristic Sequence Patterns
Non-methylated islands (NMIs) of DNA are genomic regions that are important for gene regulation and development. A recent study of genome-wide non-methylation data in vertebrates by Long et al. (eLife 2013;2:e00348) has shown that many experimentally identified non-methylated regions do not overlap with classically defined CpG islands which are computationally predicted using simple DNA sequence features. This is especially true in cold-blooded vertebrates such as Danio rerio (zebrafish). In order to investigate how predictive DNA sequence is of a region’s methylation status, we applied a supervised learning approach using a spectrum kernel support vector machine, to see if a more complex model and supervised learning can be used to improve non-methylated island prediction and to understand the sequence properties of these regions. We demonstrate that DNA sequence is highly predictive of methylation status, and that in contrast to existing CpG island prediction methods our method is able to provide more useful predictions of NMIs genome-wide in all vertebrate organisms that were studied. Our results also show that in cold-blooded vertebrates (Anolis carolinensis, Xenopus tropicalis and Danio rerio) where genome-wide classical CpG island predictions consist primarily of false positives, longer primarily AT-rich DNA sequence features are able to identify these regions much more accurately. DNA methylation is one of the most heavily studied epigenetic modifications, with active research into the topic taking place for over 30 years. Over that time a relationship between non-methylated stretches of genomic DNA and the regulation of nearby genes has been identified, and it was noted that approximately half of all human genes have an unexpectedly high number of unmethylated CpG dinucleotides in their promoters. Nevertheless, until relatively recently we haven’t been able to directly measure methylation gnome-wide, leading to the widespread use of CpG-rich regions (or CpG Islands) as a proxy for unmethylated genomic regions. Fortunately, recently developed experiments finally allow us to look at DNA methylation genome-wide, and have shown that CpG islands do not overlap true non-methylated islands to the extent we would have hoped, especially in cold-blooded vertebrates. We show that by using these experimentally determined non-methylated regions as training data, we are able to improve the prediction of non-methylated regions genome-wide in comparison to classical CpG island prediction methods, in both warm-blooded and cold-blooded vertebrates. Additionally, we show that while CpG content is an important predictor of methylation status in warm-blooded vertebrates, in cold-blooded vertebrates higher predictive accuracy can be achieved using longer stretches of relatively AT-rich DNA sequence.
- Open Access
- 28 November 2016
- Published by Oxford University Press (OUP) in Nucleic Acids Research
- Vol. 45 (4) , 1805-1819
Role of the chromatin landscape and sequence in determining cell type-specific genomic glucocorticoid receptor binding and gene regulation
The genomic loci bound by the glucocorticoid receptor (GR), a hormone-activated transcription factor, show little overlap between cell types. To study the role of chromatin and sequence in specifying where GR binds, we used Bayesian modeling within the universe of accessible chromatin. Taken together, our results uncovered that although GR preferentially binds accessible chromatin, its binding is biased against accessible chromatin located at promoter regions. This bias can only be explained partially by the presence of fewer GR recognition sequences, arguing for the existence of additional mechanisms that interfere with GR binding at promoters. Therefore, we tested the role of H3K9ac, the chromatin feature with the strongest negative association with GR binding, but found that this correlation does not reflect a causative link. Finally, we find a higher percentage of promoter-proximal GR binding for genes regulated by GR across cell types than for cell type-specific target genes. Given that GR almost exclusively binds accessible chromatin, we propose that cell type-specific regulation by GR preferentially occurs via distal enhancers, whose chromatin accessibility is typically cell type-specific, whereas ubiquitous target gene regulation is more likely to result from binding to promoter regions, which are often accessible regardless of cell type examined.
- 28 November 2016
- Published by Oxford University Press (OUP) in Nucleic Acids Research
- Vol. 45 (4) , 1805-1819
Open AccessRole of the chromatin landscape and sequence in determining cell type-specific genomic glucocorticoid receptor binding and gene regulation
The genomic loci bound by the glucocorticoid receptor (GR), a hormone-activated transcription factor, show little overlap between cell types. To study the role of chromatin and sequence in specifying where GR binds, we used Bayesian modeling within the universe of accessible chromatin. Taken together, our results uncovered that although GR preferentially binds accessible chromatin, its binding is biased against accessible chromatin located at promoter regions. This bias can only be explained partially by the presence of fewer GR recognition sequences, arguing for the existence of additional mechanisms that interfere with GR binding at promoters. Therefore, we tested the role of H3K9ac, the chromatin feature with the strongest negative association with GR binding, but found that this correlation does not reflect a causative link. Finally, we find a higher percentage of promoter-proximal GR binding for genes regulated by GR across cell types than for cell type-specific target genes. Given that GR almost exclusively binds accessible chromatin, we propose that cell type-specific regulation by GR preferentially occurs via distal enhancers, whose chromatin accessibility is typically cell type-specific, whereas ubiquitous target gene regulation is more likely to result from binding to promoter regions, which are often accessible regardless of cell type examined.
- Open Access
- 1 September 2016
- Published by PeerJ in PeerJ Preprints
Predicting enhancers using a small subset of high confidence examples and co-training
Enhancers are important regulatory regions located throughout the genome, primarily in non-coding regions. Several experimental methods have been developed over the last several years to identify their location, but the search space is large and the overlap between the putative enhancer identified using these methods tends to be very small. Computational methods for enhancer prediction often use one large set of experimentally identified enhancer regions as input, and therefore rely critically on their correctness. We chose to take a different approach, and start with a high confidence set of 21 enhancer that are in the intersection of enhancers identified using three completely unrelated experimental approaches: deepCAGE, HiCap and classical enhancer reporter assays. Because this starting set is so small, we use a semi-supervised approach called co-training rather than a fully supervised approach to progressively predict enhancers from unlabeled regions. Using this approach we are able to outperform supervised learning as well as simpler semi-supervised learning methods and achieve an average area under the ROC curve of 0.84.
- 1 September 2016
- Published by PeerJ in PeerJ Preprints
Open AccessPredicting enhancers using a small subset of high confidence examples and co-training
Enhancers are important regulatory regions located throughout the genome, primarily in non-coding regions. Several experimental methods have been developed over the last several years to identify their location, but the search space is large and the overlap between the putative enhancer identified using these methods tends to be very small. Computational methods for enhancer prediction often use one large set of experimentally identified enhancer regions as input, and therefore rely critically on their correctness. We chose to take a different approach, and start with a high confidence set of 21 enhancer that are in the intersection of enhancers identified using three completely unrelated experimental approaches: deepCAGE, HiCap and classical enhancer reporter assays. Because this starting set is so small, we use a semi-supervised approach called co-training rather than a fully supervised approach to progressively predict enhancers from unlabeled regions. Using this approach we are able to outperform supervised learning as well as simpler semi-supervised learning methods and achieve an average area under the ROC curve of 0.84.
- Open Access
- 1 September 2016
- Published by PeerJ in PeerJ Preprints
Predicting enhancers using a small subset of high confidence examples and co-training
Enhancers are important regulatory regions located throughout the genome, primarily in non-coding regions. Several experimental methods have been developed over the last several years to identify their location, but the search space is large and the overlap between the putative enhancer identified using these methods tends to be very small. Computational methods for enhancer prediction often use one large set of experimentally identified enhancer regions as input, and therefore rely critically on their correctness. We chose to take a different approach, and start with a high confidence set of 21 enhancer that are in the intersection of enhancers identified using three completely unrelated experimental approaches: deepCAGE, HiCap and classical enhancer reporter assays. Because this starting set is so small, we use a semi-supervised approach called co-training rather than a fully supervised approach to progressively predict enhancers from unlabeled regions. Using this approach we are able to outperform supervised learning as well as simpler semi-supervised learning methods and achieve an average area under the ROC curve of 0.84.
- 1 September 2016
- Published by PeerJ in PeerJ Preprints
Open AccessPredicting enhancers using a small subset of high confidence examples and co-training
Enhancers are important regulatory regions located throughout the genome, primarily in non-coding regions. Several experimental methods have been developed over the last several years to identify their location, but the search space is large and the overlap between the putative enhancer identified using these methods tends to be very small. Computational methods for enhancer prediction often use one large set of experimentally identified enhancer regions as input, and therefore rely critically on their correctness. We chose to take a different approach, and start with a high confidence set of 21 enhancer that are in the intersection of enhancers identified using three completely unrelated experimental approaches: deepCAGE, HiCap and classical enhancer reporter assays. Because this starting set is so small, we use a semi-supervised approach called co-training rather than a fully supervised approach to progressively predict enhancers from unlabeled regions. Using this approach we are able to outperform supervised learning as well as simpler semi-supervised learning methods and achieve an average area under the ROC curve of 0.84.
- 21 December 2015
- Published by Springer Nature in Oncogene
- Vol. 35 (46) , 5942-5952
SPON2, a newly identified target gene of MACC1, drives colorectal cancer metastasis in mice and is prognostic for colorectal cancer patient survival
MACC1 (metastasis associated in colon cancer 1) is a prognostic biomarker for tumor progression, metastasis and survival of a variety of solid cancers including colorectal cancer (CRC). Here we aimed to identify the MACC1-induced transcriptome and key players mediating the MACC1-induced effects in CRC. We performed microarray analyses using CRC cells ectopically overexpressing MACC1. We identified more than 1300 genes at least twofold differentially expressed, including the gene SPON2 (Spondin 2) as 90-fold upregulated transcriptional target of MACC1. MACC1-dependent SPON2 expression regulation was validated on mRNA and protein levels in MACC1 high (endogenously or ectopically) and low (endogenously or by knockdown) expressing cells. Chromatin immunoprecipitation analysis demonstrated the binding of MACC1 to the gene promoter of SPON2. In cell culture, ectopic SPON2 overexpression induced cell viability, migration, invasion and colony formation in endogenously MACC1 and SPON2 low expressing cells, whereas SPON2 knockdown reduced proliferative, migratory and invasive abilities in CRC cells with high endogenous MACC1 and SPON2 expression. In intrasplenically transplanted NOD/SCID mice, metastasis induction was analyzed with control or SPON2-overexpressing CRC cells. Tumors with SPON2 overexpression induced liver metastasis (vs control animals without any metastases, P=0.0036). In CRC patients, SPON2 expression was determined in primary tumors (stages I–III), and survival time was analyzed by Kaplan–Meier method. CRC patients with high SPON2 expressing primary tumors demonstrated 8 months shorter metastasis-free survival (MFS) compared with patients with low SPON2 levels (P=0.053). Combining high levels of SPON2 and MACC1 improved the identification of high-risk patients with a 20-month shorter MFS vs patients with low biomarker expression. In summary, SPON2 is a transcriptional target of the metastasis gene MACC1. SPON2 induces cell motility in vitro and CRC metastasis in mice. In patients, SPON2 serves as prognostic indicator for CRC metastasis and survival, and might represent a promising target for therapeutic approaches.
- 21 December 2015
- Published by Springer Nature in Oncogene
- Vol. 35 (46) , 5942-5952
SPON2, a newly identified target gene of MACC1, drives colorectal cancer metastasis in mice and is prognostic for colorectal cancer patient survival
MACC1 (metastasis associated in colon cancer 1) is a prognostic biomarker for tumor progression, metastasis and survival of a variety of solid cancers including colorectal cancer (CRC). Here we aimed to identify the MACC1-induced transcriptome and key players mediating the MACC1-induced effects in CRC. We performed microarray analyses using CRC cells ectopically overexpressing MACC1. We identified more than 1300 genes at least twofold differentially expressed, including the gene SPON2 (Spondin 2) as 90-fold upregulated transcriptional target of MACC1. MACC1-dependent SPON2 expression regulation was validated on mRNA and protein levels in MACC1 high (endogenously or ectopically) and low (endogenously or by knockdown) expressing cells. Chromatin immunoprecipitation analysis demonstrated the binding of MACC1 to the gene promoter of SPON2. In cell culture, ectopic SPON2 overexpression induced cell viability, migration, invasion and colony formation in endogenously MACC1 and SPON2 low expressing cells, whereas SPON2 knockdown reduced proliferative, migratory and invasive abilities in CRC cells with high endogenous MACC1 and SPON2 expression. In intrasplenically transplanted NOD/SCID mice, metastasis induction was analyzed with control or SPON2-overexpressing CRC cells. Tumors with SPON2 overexpression induced liver metastasis (vs control animals without any metastases, P=0.0036). In CRC patients, SPON2 expression was determined in primary tumors (stages I–III), and survival time was analyzed by Kaplan–Meier method. CRC patients with high SPON2 expressing primary tumors demonstrated 8 months shorter metastasis-free survival (MFS) compared with patients with low SPON2 levels (P=0.053). Combining high levels of SPON2 and MACC1 improved the identification of high-risk patients with a 20-month shorter MFS vs patients with low biomarker expression. In summary, SPON2 is a transcriptional target of the metastasis gene MACC1. SPON2 induces cell motility in vitro and CRC metastasis in mice. In patients, SPON2 serves as prognostic indicator for CRC metastasis and survival, and might represent a promising target for therapeutic approaches.